I remember us arguing about this for a while on IRC before 1.10 was released. IIRC the consensus back then was that we want to have public API to get the backing surface, but I think we couldn't agree on a name, so we didn't do it.

Would something like
cairo_surface_t *cairo_subsurface_get_backing_surface(cairo_surace_t *subsurface);
do what you want?

Yes, the cairo_subsurface_get_backing_surface() would be great.

I think the following commit is relevant here:

commit ba855a12e8d686f2137f82d317791f3ec4a68fc0
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Fri Oct 14 16:01:27 2011 +0100

    xlib-xp

diff --git a/src/cairo-surface-subsurface.c b/src/cairo-surface-subsurface.c
index 48d10fb..b93604f 100644
--- a/src/cairo-surface-subsurface.c
+++ b/src/cairo-surface-subsurface.c
@@ -577,6 +577,7 @@ cairo_surface_create_for_rectangle (cairo_surface_t *target,
     }
 
     surface->target = cairo_surface_reference (target);
+    surface->base.type = surface->target->type;
 
     return &surface->base;
 }
@@ -609,6 +610,7 @@ _cairo_surface_create_for_rectangle_int (cairo_surface_t *target,
     surface->extents.x += target->device_transform.x0;
     surface->extents.y += target->device_transform.y0;
     surface->target = cairo_surface_reference (target);
+    surface->base.type = surface->target->type;
 
     return &surface->base;
 }

Looks like I accidentally fixed this as every public surface should have a public type, and was already established convention that Cairo masked the internal proxies such as paginated surfaces and reported the underlying surface type instead.

We are missing a couple of getters for retrieving the target surface if you happen to have a good thesaurus handy. However, the original bug is now resolved...